Ramish Rasool

AI & Full-Stack Engineer specializing in LLM Systems, Document AI, and Cloud Platforms.

Based in: Lahore, PakistanCurrently: Lead Full-Stack Engineer & Architect, ADASTRA TUITIONExperience: 4 years

Hire me

A versatile engineer with a proven track record of architecting multi-tenant platforms and high-accuracy AI parsing systems across AWS and GCP. He has successfully led technical migrations and built generative AI backends from scratch, significantly reducing operational costs while scaling production workloads.

Experience

Lead Full-Stack Engineer & Architect

ADASTRA TUITIONUK (Remote)workJul 2026 – Present

Summary

Sole engineer designing and building Adastra Core, a multi-tenant platform (organisation → brand → centre) that runs every brand and franchise of a UK tuition group; shipped 109 API endpoints across 16 modules and 39 data models in ten weeks, with three apps live on AWS.

What I did

  • Leading the migration of a live business from Supabase to AWS: replaced 55 row-level-security policies with API-level tenancy and wrote a 25-case cross-tenant IDOR test suite that caught a real access-control gap before launch.
  • Built an idempotent ETL with reconciliation across 23 source tables (zero discrepancies on rehearsal) and a cutover runbook with 12 go/no-go criteria and a 13-step rollback to protect live bookings and payments.
  • Provisioned the full AWS estate (ECS Fargate, RDS, Amplify) as 62 Terraform resources with push-to-deploy GitHub Actions over OIDC and no long-lived keys; dev environment runs at ~$41/month. Moved from part-time to full-time on early delivery.
  • Configured the API to dynamically create Stripe Checkout sessions based on the organization owning the specific center.
  • Designed the webhook handler to verify signatures against organization-specific signing secrets.
  • Architected infrastructure including NestJS on ECS Fargate, ALB, Amplify frontends, S3 with presigned uploads, and Cloudflare DNS.
  • Configured scoped deployment roles and automated Prisma migrations as one-off ECS tasks.

Results

  • Architected a multi-tenant payment system where each organization in the hierarchy manages its own Stripe account.
  • Secured organization-specific Stripe keys using AWS Secrets Manager instead of the database or environment configuration.
  • Implemented a server-side logic to determine Stripe test or live mode, preventing clients from switching booking modes.
  • Developed a reliable booking confirmation system using verified webhooks and idempotent event tracking to prevent duplicate processing.
  • Managed a live business migration to Stripe including go/no-go checks and rollback procedures.
  • Wrote the entire AWS estate in Terraform, managing approximately 62 resources from an empty account.
  • Configured a VPC with RDS Postgres in isolated subnets and no public database endpoint.
  • Implemented database access via an outbound-only EC2 host using SSM port-forwarding.
  • Established CI using GitHub Actions with OIDC to eliminate long-lived AWS keys.
  • Resolved a database connection failure by implementing a least-privilege database role for the application.
  • Fixed a pipeline change-detection bug that skipped backend deploys on multi-commit pushes.
  • Caught a real access-control gap before launch using a 25-case cross-tenant IDOR test suite.
  • Achieved zero discrepancies on ETL rehearsal across 23 source tables.
  • Reduced dev environment costs to ~$41/month.
TypeScriptNestJSNext.jsPostgreSQLAWSTerraformECS FargateRDSAmplifyGitHub ActionsOIDCSupabaseStripe WebhooksStripe CheckoutAmazon ECS

AI & Machine Learning Instructor

LUMS CENTRE FOR CONTINUING EDUCATION (CES)Lahore, PakistanworkApr 2026 – May 2026

Summary

Designed and taught a 6-week, 12-class applied AI/ML course from scratch for 45 professionals across 9 industries, building live demos such as an automated call-quality evaluator for a 500-agent BPO.

Results

  • Taught 45 professionals across 9 industries.
  • Built an automated call-quality evaluator for a 500-agent BPO.
AIMachine Learning

Senior Software/Machine Learning Engineer

MERCHANT MATRIXLos Angeles, US (Remote)workOct 2024 – Mar 2026

Summary

Architected a template-first hybrid parser (regex templates for 20+ major US bank formats, Gemini fallback) with a balance-reconciliation check; raised zero-discrepancy extraction from ~40% to 92% and cut AI API costs 40%+ (~$0.48 per statement all-in). Built the parallel email-intake pipeline (async queues, exponential-backoff retries for Google API rate limits) that has processed 10,000+ bank statements with a 100/day peak, cutting manual review from ~30 minutes to seconds per statement. Proposed and led the migration from DigitalOcean to Google Cloud (Cloud Run, Cloud SQL, Pub/Sub), parallelizing document processing to remove memory bottlenecks under high volume. Engineered an AI underwriting assistant using RAG over funder guidelines and Gemini function calling to generate "Best Offer" recommendations and auto-declines from computed risk signals (negative days, daily balances). One of two engineers working directly with the founders; shipped 40+ CRM features including deal management, billing, notifications and a KPI dashboard.

What I did

  • Used a reconciliation check to measure extraction accuracy improvements.

Results

  • Proposed and led the infrastructure migration from DigitalOcean to GCP.
  • Parallelized bank statement processing through queues to resolve memory pressure issues.
  • Cut manual review from ~30 minutes to seconds per statement.
  • Shipped 40+ CRM features.
PythonFastAPINext.jsGeminiRAGGCPPostgreSQLCloud RunCloud SQLPub/SubDigitalOceanGoogle API

Skills

Technical

FastAPIFastAPI
Function Calling
LoRA fine-tuning
Stable Diffusion
Gemini
GPT
Llama
RAG
NestJS
Node.jsNode.js
Next.jsNext.js
PostgreSQLPostgreSQL
AWS
GCPGCP
TerraformTerraform
DockerDocker
Celery
GitHub ActionsGitHub Actions
Runpod
ComfyUI
pgvector
Claude
Hugging Face
PyTorchPyTorch
ReactReact
Prisma
RedisRedis
Stripe
Google Cloud Pub/SubGoogle Cloud Pub/Sub
Stripe Connect
Model Context Protocol (MCP)
Stripe Billing
Model Evaluation
SwiftSwift
ScreenCaptureKit
Electron
Deepgram
AWS Systems Manager (SSM)

Languages

JavaScript
TypeScript
Python
SQL
Go
Shell
C++
Kotlin
Java
Haskell

Projects

Codebase-specific Coding Assistant

Lead Engineerprofessional
Fine-tuned DeepSeek R1 into a codebase-specific coding assistant for a client: data preparation, LoRA training on cloud GPUs (PyTorch, Hugging Face, W&B), evaluation against the client's test questions, and deployment on Hugging Face Spaces and Runpod.
DeepSeek R1LoRAPyTorchHugging FaceW&BHugging Face SpacesRunpod

NFT Marketplace on Custom Blockchain

Team Leadacademic
Built an NFT marketplace on a blockchain implemented from scratch. Placed 1st of 64 university teams nationally in a blockchain hackathon.
BlockchainNFT

High-performance serving layer at LAAM

Developerpersonal
Built a high-performance serving layer for a fashion marketplace's main web page to handle high traffic and concurrent requests.
GoElasticsearch

Multi-server chat application

personal
Developed a systems-level chat application using sockets, multithreading, and two-phase commit during CS degree at LUMS.
C++

Tacit

Sole Creatorpersonal
A macOS productivity application that integrates live meeting transcription with task management. It generates briefings from Linear issues, transcribes system and microphone audio, and uses a Claude agent to automatically update task trackers based on meeting discussions.
SwiftScreenCaptureKitElectronTypeScriptDeepgramClaude

Education

Bachelor of Science

LAHORE UNIVERSITY OF MANAGEMENT SCIENCES (LUMS)Computer ScienceSep 2018 – Jun 2022

Recognition

Awards

Dean's Honor List

Lahore University of Management Sciences (LUMS)2020

Awarded for 2 consecutive years (2019, 2020) for maintaining a CGPA greater than 3.6

1st Place - National Blockchain Hackathon

National Blockchain Hackathon

Placed 1st of 64 university teams nationally, leading a 3-person team that built an NFT marketplace on a blockchain implemented from scratch.

Hire RamishReach out about a role, a contract or a conversation.For recruiters

Ramish's twin is AI, it can make mistakes.